Recognition of EEG based on Improved Black Widow Algorithm optimized SVM

As a classifier suitable for nonlinear samples, support vector machines (SVM) is widely used in Electroencephalogram (EEG) signals pattern recognition. The performance of SVM depends mainly on the selection of model parameters. This paper modifies the reproductive formula of the Black Widow Optimization (BWO) algorithm to obtain an improved Black Widow Optimization (IBWO) algorithm, which solves the problem of the restricted search radius of the BWO algorithm. We construct an IBWO-SVM model in which IBWO is used to optimize the penalty parameter C and the kernel function parameter g of the SVM, and the optimized SVM model is applied to EEG signals classification. The effectiveness and superiority of IBWO-SVM is verified by using the BCI competition III dataset IVa EEG public dataset, and compared with the SVM optimized by the original BWO algorithm (BWO-SVM), the artificial fish swarm algorithm (AFSA-SVM) and the particle swarm optimization algorithm (PSO-SVM), The simulation results demonstrate that all indicators of the IBWO-SVM model are better than the three comparison models, in which the average classification accuracy on five subjects reaches 97.29%. In addition, the two-class motor imagery (left arm stretch and right fist clenched) EEG data collected from our experiment is also used to test the performance of the IBWO-SVM model. The average classification accuracy of the final twelve subjects is 5.21%, 4.16% and 1.47% higher than that of BWO-SVM, AFSA-SVM, and PSO-SVM, respectively. Therefore, the IBWO-SVM model can effectively improve the accuracy of EEG signals pattern recognition, which has a very important practical value.